The Relationship between Poverty and Healthcare Seeking ... · Charles Darwin University The Relationship between Poverty and Healthcare Seeking among Patients Hospitalized with Acute

Charles Darwin University

The Relationship between Poverty and Healthcare Seeking among PatientsHospitalized with Acute Febrile Illnesses in Chittagong, Bangladesh

Herdman, M. Trent; Maude, Richard; Chowdhury, Safiqul; Kingston, Hugh William Fluellen;Jeeyapant, Atthanee; Samad, Rasheda; Karim, Rezaul; Dondorp, Arjen; Hossain, AmirPublished in:PLoS One

DOI:10.1371/journal.pone.0152965

Published: 01/01/2016

Document VersionPublisher's PDF, also known as Version of record

Link to publication

Citation for published version (APA):Herdman, M. T., Maude, R., Chowdhury, S., Kingston, H. W. F., Jeeyapant, A., Samad, R., Karim, R., Dondorp,A., & Hossain, A. (2016). The Relationship between Poverty and Healthcare Seeking among PatientsHospitalized with Acute Febrile Illnesses in Chittagong, Bangladesh. PLoS One, 11(4), 1-21.https://doi.org/10.1371/journal.pone.0152965

General rightsCopyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright ownersand it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.

• Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain • You may freely distribute the URL identifying the publication in the public portal

Take down policyIf you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediatelyand investigate your claim.

Download date: 14. Dec. 2020

https://doi.org/10.1371/journal.pone.0152965

https://researchers.cdu.edu.au/en/publications/7c4b7907-0ba5-467c-a98d-cffaba020b7c

https://doi.org/10.1371/journal.pone.0152965

RESEARCH ARTICLE

The Relationship between Poverty andHealthcare Seeking among PatientsHospitalized with Acute Febrile Illnesses inChittagong, BangladeshM. Trent Herdman1,2*, Richard James Maude1,3, Md. Safiqul Chowdhury4, HughW.F. Kingston1,5, Atthanee Jeeyapant1, Rasheda Samad4, Rezaul Karim4, ArjenM. Dondorp1,3, Md. Amir Hossain4

1 Mahidol-Oxford Tropical Medicine Research Unit, Faculty of Tropical Medicine, Mahidol University,Bangkok, Thailand, 2 University College, University of Oxford, Oxford, United Kingdom, 3 Centre forTropical Medicine, Nuffield Department of Clinical Medicine, Churchill Hospital, Oxford, United Kingdom,4 Chittagong Medical College Hospital, Chittagong, Bangladesh, 5 Global Health Division, Menzies Schoolof Health Research and Charles Darwin University, Darwin, Northern Territory, Australia

*[email protected]

AbstractDelays in seeking appropriate healthcare can increase the case fatality of acute febrile ill-

nesses, and circuitous routes of care-seeking can have a catastrophic financial impact

upon patients in low-income settings. To investigate the relationship between poverty and

pre-hospital delays for patients with acute febrile illnesses, we recruited a cross-sectional,

convenience sample of 527 acutely ill adults and children aged over 6 months, with a docu-

mented fever�38.0°C and symptoms of up to 14 days’ duration, presenting to a tertiary

referral hospital in Chittagong, Bangladesh, over the course of one year from September

2011 to September 2012. Participants were classified according to the socioeconomic sta-

tus of their households, defined by the Oxford Poverty and Human Development Initiative’s

multidimensional poverty index (MPI). 51% of participants were classified as multidimen-

sionally poor (MPI>0.33). Median time from onset of any symptoms to arrival at hospital

was 22 hours longer for MPI poor adults compared to non-poor adults (123 vs. 101 hours)

rising to a difference of 26 hours with adjustment in a multivariate regression model (95%

confidence interval 7 to 46 hours; P = 0.009). There was no difference in delays for children

from poor and non-poor households (97 vs. 119 hours; P = 0.394). Case fatality was 5.9%

vs. 0.8% in poor and non-poor individuals respectively (P = 0.001)—5.1% vs. 0.0% for poor

and non-poor adults (P = 0.010) and 6.4% vs. 1.8% for poor and non-poor children (P =

0.083). Deaths were attributed to central nervous system infection (11), malaria (3), urinary

tract infection (2), gastrointestinal infection (1) and undifferentiated sepsis (1). Both poor

and non-poor households relied predominantly upon the (often informal) private sector for

medical advice before reaching the referral hospital, but MPI poor participants were less

likely to have consulted a qualified doctor. Poor participants were more likely to attribute

delays in decision-making and travel to a lack of money (P<0.001), and more likely to face

PLOSONE | DOI:10.1371/journal.pone.0152965 April 7, 2016 1 / 21

OPEN ACCESS

Citation: Herdman MT, Maude RJ, Chowdhury M.S,Kingston HWF, Jeeyapant A, Samad R, et al. (2016)The Relationship between Poverty and HealthcareSeeking among Patients Hospitalized with AcuteFebrile Illnesses in Chittagong, Bangladesh. PLoSONE 11(4): e0152965. doi:10.1371/journal.pone.0152965

Editor: Mohammad Ali, Johns Hopkins BloombergSchool of Public Health, UNITED STATES

Received: August 17, 2015

Accepted: March 22, 2016

Published: April 7, 2016

Copyright: © 2016 Herdman et al. This is an openaccess article distributed under the terms of theCreative Commons Attribution License, which permitsunrestricted use, distribution, and reproduction in anymedium, provided the original author and source arecredited.

Data Availability Statement: All relevant data areavailable from the Dryad database (DOI: 10.5061/dryad.4f6fd).

Funding: MTH was funded by the RadcliffeTravelling Fellowship, University College, Universityof Oxford. The funders had no role in study design,data collection and analysis, decision to publish, orpreparation of the manuscript.

Competing Interests: The authors have declaredthat no competing interests exist.

http://crossmark.crossref.org/dialog/?doi=10.1371/journal.pone.0152965&domain=pdf

http://creativecommons.org/licenses/by/4.0/

http://dx.doi.org/10.5061/dryad.4f6fd

http://dx.doi.org/10.5061/dryad.4f6fd

catastrophic expenditure of more than 25% of monthly household income (P<0.001). We

conclude that multidimensional poverty is associated with greater pre-hospital delays and

expenditure in this setting. Closer links between health and development agendas could

address these consequences of poverty and streamline access to adequate healthcare.

IntroductionBangladesh, in common with most rapidly developing countries, is subject to profound inequi-ties of wealth and health [1–4]. Poverty—the state of multidimensional deprivation in whichbasic needs cannot be met—is inextricably linked with disease. This relationship is complexand bi-directional: at its worst, it yields a vicious cycle of deprivation leading to disease, andcosts of illness leading to further impoverishment [5]. The goals of alleviating poverty andimproving global health have become a converging focus of development and public health ini-tiatives nationally and internationally [6–9]. Reducing inequities of wealth and healthcare hasan essential role in addressing the burden of many diseases, and quantifying these inequities isa fundamental prerequisite [7].

Acute febrile illness (AFI) accounts for the majority of illness episodes in the ChittagongDivision, and for much of the excess burden of disease associated with poverty [10, 11]. Inmost cases, the aetiology of AFI is unknown at the time of admission to the referral hospital;once hospitalized, characterization of the disease usually remains limited by available diagnos-tics and cost to clinical impression from symptoms, signs, basic microbiology, and malariadiagnostics. This aetiologic uncertainty is a major challenge to effective clinical care and publichealth, and necessitates the pragmatic approach of a broad case definition when consideringhealthcare-seeking behaviour [12].

Prompt and effective treatment of malaria, meningitis, enteric fever, sepsis, and other causesof serious AFI in this setting can save lives and reduce morbidity [13–17]. Hence, identifyingand addressing barriers to care is essential. However, there is limited understanding of thesocioeconomic risk factors and consequences of AFI. Most reports on the correlation betweenAFI and poverty have come from community-based surveys, where the majority of illnessesencountered are self-limiting and minimally investigated [10, 18]. In contrast, patients with themost burdensome and best-characterized infections converge upon the in-patient hospital set-ting, where reports of morbidity and mortality are frequently compiled, but rarely disaggre-gated by socioeconomic status.

Newly validated tools—with concise and robust parameters of assessment—facilitate assess-ment of a patient’s exposure to poverty in the context of an acute illness [18, 19]. The multidi-mensional poverty index (MPI) was developed by the Oxford Poverty and HumanDevelopment Initiative (OPHI) with the aim of providing a validated, easily administered, andinternationally applicable metric for assessing household deprivation, and steer recommenda-tions to reduce poverty [20]. This index identifies household living standards, education, andchronic health status (defined by nutritional status and exposure to child mortality) as co-exist-ing dimensions of poverty, and links its assessment parameters directly to the priorities of theMillennium Development Goals. The United Nations Development Programme has recentlyadopted MPI as an international standard for assessment, tracking, and planning of progress inthe global fight against poverty [21].

This investigation seeks to complement previous, community-based studies of the socioeco-nomic background of people with AFI in Bangladesh, by characterizing the subset of patientsadmitted for acute medical management [10, 11, 22–24]. We report a survey of patients with

Poverty and Healthcare-Seeking in Acute Febrile Illnesses

PLOS ONE | DOI:10.1371/journal.pone.0152965 April 7, 2016 2 / 21

AFI attending a large referral hospital in Bangladesh, and describe the relationship betweenpoverty and pre-hospital delays.

Materials and Methods

Local structure of healthcare and referral networkChittagong Division is the largest of Bangladesh’s eight administrative Divisions, with a popu-lation of approximately 27 million in 2011, at the time of this study [25]. Public-sector health-care is the responsibility of Union Health Centers (which provide outreach services focused onprevention) and Upazila/Thana Health Complexes. These Government Health Complexes(GHC) are intended to provide a broad range of out-patient services, and have very limiteddiagnostic facilities (such as rapid diagnostic tests, RDTs); most also support 30–50 in-patientbeds under the supervision of a small medical and nursing team. Secondary level services areprovided by District Hospitals, with out-patient facilities, 50 to 250 in-patient beds, and limitedlaboratory and radiographic capabilities. Within the public sector, consultations with health-care workers are free of charge, but fees for provision of medication and investigations, as wellas inpatient care, vary. Health Complexes and District Hospitals both make direct referrals totertiary referral hospitals such as Chittagong Medical College Hospital (CMCH), where thisinvestigation was undertaken [26, 27]. The true catchment population of CMCH is difficult todefine. In addition to formal referrals from public and private secondary level services, a largenumber of patients are admitted via the Emergency Department after attending on the infor-mal advice of practitioners or by self-referral.

Alongside public sector health facilities, the private sector delivers a large proportion of medi-cal care at all levels, where payment for consultations, investigations, and treatment is usuallyout-of-pocket. It is estimated that only 10% of all health and family planning consultations occurin the government sector in Bangladesh [28]. Shops and pharmacies sell over-the-counter andprescription medication, and many shopkeepers and pharmacists give informal medical advice.A spectrum of practitioners operates private chambers for fee-based out-patient consultations,including licensed specialists, other officially qualified para-professionals, and ‘village doctors’who lack formal qualifications but provide allopathic advice and treatment [18, 27, 28]. For thepresent investigation, we define doctors as the holders of an MBBS (Medical Bachelor/Bachelorof Surgery), LMF (Licentiate of the State Medical Facility), or higher professional qualification.We define Allopathic Practitioners as those who provide allopathic healthcare advice in a privatechamber, but who lack MBBS, LMF, or higher qualification (or whose qualification is unknown).Alongside Allopathic Practitioners, healers from homoeopathic, herbalist, Ayurvedic, and spiri-tual backgrounds also provide health advice and treatment within the private sector, and are heredefined separately, as Traditional Healers [10, 22]. In-patient services are also present in the pri-vate sector, with numerous private hospitals, concentrated in urban centers.

Study siteChittagong Medical College Hospital (CMCH) is the principal public-sector referral hospitalfor Chittagong Division. Participants were recruited continuously from September 2011 toSeptember 2012. Ethical approval for this study was obtained from the CMCH Ethical ReviewCommittee and the Oxford Tropical Research Ethics Committee.

The hospital has 1,313 beds, and usually operates at much greater than 100% occupancy.Patients were recruited from the three adult general medical wards and one general pediatricward. Over the study period, a total of 39,077 patients were admitted to the adult medicalwards, and 15,514 to the pediatric ward with all clinical presentations; the total number ofpatients presenting with AFI was not available.



Screening and recruitment proceduresInformed, written consent was obtained from patients or legally acceptable representatives inall cases. For adults with capacity to give consent to participate, informed, written consent wasobtained from the patient directly. For children and adults without capacity to give consent,informed, written consent was obtained on behalf of the patient from the next of kin, caretak-ers, or guardians.

A team of six medical and pediatric resident junior doctors acted as interviewers for this sur-vey. All interviewers were fluent speakers of Bengali and Chittagonian. Interviewers receivedtraining in Good Clinical Practice for Research, interview techniques, and standard operatingprocedures for recruitment and use of the survey and anthropometric measurement tools. Thetarget sample size of approximately 500 participants over one year was determined based onthe estimated capacity of the interviewers to balance study procedures with their full-time clini-cal duties. Patients admitted with acute febrile illnesses were identified for screening throughdaily liaison with the clinical teams responsible for ward admissions. Sampling was structuredover time by the minimum target of daily recruitment of one adult and one paediatric patient,but on a given day, if multiple patients were eligible, a convenience sample was taken. Inter-views and anthropometric measurements were conducted at the bedside with patients. Toassist participants with information recall and obtaining heights, weights, and mid upper armcircumferences, other household members were encouraged to remain at the bedside duringthe interview and contribute information, provided they and the participant gave verbal con-sent for them to remain.

Participant eligibility was dependent upon consent, an age of greater than six months, a doc-umented fever of greater than or equal to 38.0°C within 48 hours of admission to a medical orpediatric ward, and a history of symptoms of no greater than 14 days’ duration. Patients andhousehold members aged under 18 years were classified as children, in keeping with the UnitedNations’ definition, to maintain consistent classification for clinical, occupational, and medico-legal purposes. In keeping with the definition used during Demographic and Health Surveys(DHS) data collection, we regarded all people who usually reside and eat together as householdmembers [29].

Interview surveyParticipants completed a face-to-face, interviewer-assisted survey. A pilot survey was under-taken with 60 participants to test questions for clarity and consistency (data not shown). Pilotdata are not included in this analysis, as inclusion criteria changed during the pilot phase.

A 15-minute structured interview recorded 59 multiple-choice, yes/no, and short free-textfields, addressing the following variables: patient and household demographics and location,symptom timespan, sources of help and advice, causes of delays, mode of transport, estimatedcosts associated with the illness prior to arrival at the referral hospital, time lost from work andhousehold duties (by all working household members—patient plus carers), estimated house-hold income, and other household details needed to calculate the Multidimensional PovertyIndex (MPI), as described below.

Participants were interviewed within 24 hours of admission if possible, and followed upuntil discharge from the ward, transfer to another facility, or death, whereupon this outcomewas recorded, along with the provisional diagnosis from the clinical team.

Calculation of poverty and spending indicesMultidimensional Poverty Index (MPI) was calculated for each household according to thealgorithms of the OPHI [20]. Participants were classified on the basis of the MPI of their



household as MPI poor or MPI non-poor. In brief, ten dichotomous indicators of deprivationwere assessed. Six indicators of deprived living standards were weighted as 1/18 of the MPIscore: deprivation of electricity, household assets, floor material, cooking fuel, water source,and sanitation. Two indicators of educational deprivation were weighted as 1/4: lack of schoolenrolment, and lack of overall educational attainment. Two indicators of chronic health depri-vation were weighted as 1/4: presence of malnutrition in any household member, and theoccurrence of child mortality within living memory. Missing data were treated according toOPHI recommendations, and a poverty score for each household was calculated as the sum ofthe ten weighted indicators, to give a value between 0.00 and 1.00. Households with an MPI ofgreater than 0.33 were classified as multidimensionally poor. Intensity of poverty and MPI ofthe population as a whole were calculated as described by the OPHI [20]. National MPI valuesfor Bangladesh were obtained from the 2014 OPHI Country Briefing, based on data collectedin the 2011 DHS [2, 29].

Participants were asked to estimate income in an average month from all sources; this wasdivided by the number of adults in the household to determine income in Tk per adult equiva-lent (AE) per month. Adult equivalent income was calculated from the number of adults andchildren in the household using the OECD Equivalence Scale, as AE = 1+0.7(Nadults-1)+-0.5Nchildren [30]. Households were classified as above or below wealth thresholds calculatedfrom private individual purchasing power parity (PPP), using the World Bank’s PPP analysisin 2010, adjusted to the currency exchange rate from the start of the study period (1347Tk/AE/month for US$1.25/adult/day and 2155Tk/AE/month for US$2.00/adult/day) [31].

Participants were asked to estimate and characterize costs relating to illness incurred up tothe point of admission to hospital, and to describe how these costs were met. Estimated totalexpenditure during the illness episode prior to hospital admission was used to determine if par-ticipants had exceeded established definitions of catastrophic expenditure at 25%, 40%, and100% of total monthly household income [32].

Sequence of healthcare providers and timecourse estimationTo characterize healthcare-seeking behavior, participants were first asked to narrate the stepstaken in seeking help with the illness, listing all sources of help outside of the home, which hadbeen consulted during this illness episode, up to the point of arrival at CMCH. Interviewersthen screened for omitted sources from a list of common options. Participants were then askedabout the sequence in which these sources were consulted, based on the time of first consulta-tion with each. Repeated consultations with the same provider were scored as a single episode,in keeping with previous studies [33].

Having established the sequence of sources of help, estimates of the timecourse of the illnessand healthcare-seeking behavior were sought. Participants estimated the date and time atwhich the first symptom arose, approximating where a precise time or date could not berecalled by reference to day, night, and mealtimes. Participants estimated three further mile-stones: the time and date of the decision to seek help outside of the home (from the first sourcein the sequence of healthcare providers); the time and date of the decision to escalate care bycoming to the referral hospital; and the time and date of arrival at the hospital. These mile-stones were used to calculate total timespan of healthcare-seeking, subdivided into (i) the time-span from onset of symptoms to help outside the home; (ii) timespan from first help to thedecision to escalate to CMCH; and (iii) the timespan from this decision to arrival at thehospital.

Participants were questioned about perceived sources of delay in decision-making andtransport, screening from a list of common causes, with scope to volunteer additional answers.



Measurement of height, weight, and mid upper arm circumferenceMeasurements of standing height, weight, and left mid upper arm circumference (MUAC, cm)were obtained from patients and all available household members. Thresholds for malnutritionwere adapted from the OPHI standards used for national calculations of MPI [20]. For chil-dren, age- and sex-specific standard distributions were obtained from the World Health Orga-nization, and individuals more than two standard deviations from the mean (on the basis ofMUAC for children under five and BMI for those five and over) were classed as malnourished[34, 35]. Adults were classed as malnourished if they had BMIs of less than 18.5. For partici-pants who were unable to stand, MUAC alone was used, with a threshold of two SD below theage- and sex-specific mean for those under 18,<20 cm for men aged 18 or over, and<19 cmfor women aged 18 or over [34].

Data management and analysisDouble data entry was performed using OpenClinica Database Software v.3.1.3, with discrepan-cies and empty fields prompting review of the original case record forms for clarification and cor-rection. To cross-check accurate ascertainment in the face-to-face survey, the records of 67participants were validated with telephone follow-up to the participant from a second researcherafter discharge from hospital, confirming that key parameters had been correctly ascertained.

Statistical analysis was performed using STATA/IC software v.11.2 (StataCorp, College Sta-tion, TX, USA) and Prism v.6.0b (Graphpad Software, La Jolla, CA, USA). Rank correlationsbetween ordinal MPI score (0.00 to 1.00) and other variables were sought using Spearman’srho in view of the score’s non-Gaussian distribution. For univariate analysis, associationsbetween MPI status (poor or non-poor) and dichotomous variables were sought using propor-tion tests when all group sizes were greater than 10, and Fisher’s exact test when less than 10.Correlations between MPI status and continuous variables showing a non-Gaussian distribu-tion were sought using the Mann-Whitney U test. The relationship between pre-hospital illnesstimespan and explanatory variables was interrogated with multiple linear regression analysisusing the STATA software package. P-values of<0.05 were considered statistically significant.

Results

Demographic and clinical characteristics of the study population536 acutely febrile participants were recruited into this study; nine of these were excluded uponreview of data for failing to meet inclusion criteria, and data from the remaining 527 are presentedfor the remainder of this report (Fig 1). The study population comprised 242 adults (18 andolder) and 285 children; demographic and clinical characteristics are shown in Table 1. The agedistribution reflects that of the general population of Bangladesh, but shows a greater preponder-ance of children under five. 49% of recruitment was from the PaediatricWard, compared to 40%of all hospital admissions during the study period. 63% of participants were male, compared to53% of all hospital admissions during the study period, and 51% of the general population [36].

The diagnostic categories used in the study differed from those routinely collected by thehospital, but patterns of disease within the study appeared to correspond with the disease pro-file of AFI admissions (unpublished data, MAH). The overall case fatality in the study popula-tion was 3.4% (2.1% for adults and 4.6% for children).

Determination and validation of Multidimensional Poverty Index statusMPI results and their decomposition are shown in Table 2, alongside national statistics fromthe 2011 DHS dataset [2]. In keeping with findings of the DHS in the Chittagong Division, a



Fig 1. Study participants and admission outcomes, from September 2011 to September 2012. Limitedadmission record-keeping prevented ascertainment of the total number of eligible patients during the studyperiod.

doi:10.1371/journal.pone.0152965.g001

Table 1. Baseline characteristics of participants and households.

All MPI Poor MPI Non-Poor Poor vs. Non-Poor

Characteristic n = 527 n = 269 n = 258 P-value*

Age > 0.5 to < 5 years (%) 152 (29%) 92 (34%) 60 (23%) <0.001

5 to < 10 years (%) 69 (13%) 46 (17%) 23 (9%)

10 to < 18 years (%) 64 (12%) 33 (12%) 31 (12%)

18 to < 30 years (%) 126 (24%) 47 (17%) 79 (31%)

30 to < 40 years (%) 45 (9%) 14 (5%) 31 (12%)

40 to < 50 years (%) 29 (6%) 16 (6%) 13 (5%)

50 to < 60 years (%) 18 (3%) 10 (4%) 8 (3%)

> 60 years (%) 24 (5%) 11 (4%) 13 (5%)

Sex Male (%) 330 (63%) 168 (62%) 162 (63%) 0.936

Female (%) 197 (37%) 101 (38%) 96 (37%)

Children in Household, median [IQR, range] 2[1–3, 0–45] 2 [2–3, 0–9] 2 [1–3, 0–45] 0.003

Adults in Household, median [IQR, range] 4 [2–5, 1–23] 3 [2–5, 1–13] 4 [3–5, 1–23] 0.006

Residencea Rural (%) 296 (56%) 184 (68%) 112 (43%) <0.001

Urban (%) 23(44%) 85 (32%) 146 (57%)

Distance to CMCH in hr, median [IQR, range]b 2 [1–3, 0.1–13] 2 [1–3, 0.2–10] 1 [0.5–2, 0.1–13] <0.001

Length of Stay in days, median [IQR, range]c 5 [4–8, 0–64] 6 [4–8, 1–48] 5 [3–7, 0–64] 0.003

Mortality Overall (%) 18 (3.4%) 16 (5.9%) 2 (0.8%) 0.001

Among Children n = 285 (%) 13 (4.6%) 11 (6.4%) 2 (1.8%) 0.083

Among Adults n = 242 (%) 5 (2.1%) 5 (5.1%) 0 (0%) 0.010

*P-values represent comparisons of all poor vs. all non-poor participants. P-values are derived from Wilcoxon Rank Sum test for ordinal variables (age,

household size, and length of stay), proportion test was used for dichotomous variables with large group sizes (sex, rural/urban residence), and Fisher’s

exact test for mortality in view of small group sizes.aParticipants self-categorized households as rural or urban.bFrom 525 patients;cFrom 508 patients (deaths excluded).

doi:10.1371/journal.pone.0152965.t001



large percentage of participants had deprivation of cooking fuel, floor material, householdassets, and malnutrition. A smaller percentage of both the study population and national sur-veys were deprived in terms of access to drinking water, exposure to child mortality, and non-enrollment of school-aged children. Fig 2 shows that MPI score and average monthly incomeper adult correlate inversely, with a higher poverty index associated with a lower self-estimatedincome (P<0.001; Spearman’s rho -0.37). Using the standard MPI threshold of 0.33—indicat-ing substantial deprivation in at least one third of weighted indicators [20]—51% of partici-pants (95%CI 47–55%) came from households classified as multidimensionally poor. 41% ofadult patients and 60% of children came from households classed as MPI poor. The intensityof poverty in the study population as a whole (the mean percentage of weighted indices pres-ent) was 34% (95%CI 32–35%). Multidimensional poverty index (percentage of population liv-ing in poverty x intensity of poverty) of the study population as a whole is 0.171. The 2011DHS indicates that in Chittagong Division as a whole at the time of the study, 51% of the popu-lation were multidimensionally poor (95%CI 46–55%), with an overall intensity of poverty of50% (95%CI 47–52%) and overall MPI of 0.252 [2].

Household and disease characteristics disaggregated by MPI statusThe demographic characteristics of the study population disaggregated by MPI status areshown in Table 1. A larger proportion of participating children under 10 were fromMPI poorhouseholds, and a larger proportion of young adults were from non-poor households. The dis-proportionate ratio of male to female patients was observed consistently in both MPI poor and

Table 2. Prevalence of MPI components as % of the population, among study participants compared to the DHS 2011 Survey of ChittagongDivision.

StudyParticipants Chittagong Division (DHS 2011)

n = 527 n = 14,995

MPI Componenta % deprived (95%CI) % deprived (95%CI)

Education: Schooling 35.5 (31.4, 39.7) 17.6 (13.9, 21.3)

Child School Attendance 26.5 (22.8, 30.5) 17.2 (13.5, 20.9)

Health: Child Mortality 13.4 (10.6, 16.7) 25.0 (22.5, 27.6)

Nutrition 45.0 (41.4, 45.0) 35.2 (32.1, 38.3)

Living Standards: Electricity 23.3 (19.8, 27.2) 33.3 (26.9, 39.8)

Improved Sanitation 34.5 (30.5, 38.8) 39.5 (34.4, 44.6)

Drinking Water 9.9 (7.5, 12.7) 1.8 (1.0, 2.5)

Floor 44.2 (39.9, 48.6) 75.2 (72.0, 78.3)

Cooking Fuel 75.0 (71.1, 78.7) 87.6 (84.3, 91.0)

Asset Ownership 55.7 (51.3, 60.0) 83.1 (80.4, 85.9)

a Each component is scored dichotomously with the following criteria: schooling deprived if no member of the household has completed more than five

years of full-time education; child school attendance deprived if the household has children aged 5–13 who are not in full-time school; child mortality

exposed if, within living memory of the survey participant, any child in the household has died; nutrition deprived if any member of the household available

for measurement meets anthropometric criteria for malnutrition; electricity deprived if the household has no electricity; improved sanitation deprived if the

household’s toilet does not meet MDG standards, or is shared with other households; drinking water deprived if the household’s usual source of drinking

water does not meet the MDG standards for an improved water source, or if that source is more than 30 minutes’ journey away; floor deprived if the floor

is made of earth, sand, or dung; cooking fuel deprived if the household cooks with wood, charcoal, dung, straw, shrubs, or grass; asset ownership

deprived if the household has no car, truck, or tractor and has fewer than two items from a list of radio, television, telephone, refrigerator, bicycle, and

motorcycle.




non-poor populations. MPI poor households had more children and fewer adults. A larger pro-portion of the MPI poor population came from households in a rural setting, at a greater dis-tance from the hospital (with a median journey duration of 2 hours for poor and 1 hour fornon-poor households). These observations are consistent with known geographical and demo-graphic patterns of the Chittagong Division [10].

Length of admission was marginally greater for MPI poor participants. 98% of participantsacross both groups were admitted for at least 48 hours. Case fatality was 5.9% vs. 0.8% in poorand non-poor individuals respectively (P = 0.001)—5.1% vs. 0.0% for poor and non-pooradults (P = 0.010) and 6.4% vs. 1.8% for poor and non-poor children (P = 0.083).

The working diagnoses, recorded at the time of discharge from the ward or death, are sum-marized in Table 3. In most cases, the diagnosis was clinical, as laboratory and radiological ser-vices in the hospital are limited. CNS infections and malaria comprised a significantly higherproportion of AFIs in the MPI poor group, which were also the diseases carrying the greatestrisk of death. Enteric fever and dengue fever were clinically diagnosed in a significantly greaterproportion of the non-poor group.

Patterns of care-seeking prior to Referral Hospital admissionFig 3 and Table 4 summarize the care-seeking practices of participants; a complete record ofthis transition analysis is included as S1 and S2 Tables online. Among adults, participantsfrom poor households were more likely to consult an unqualified allopathic practitioner,

Fig 2. MPI score vs. average household income per adult equivalent per month. The correlation issignificant with a P-value of <0.001 and a Spearman’s rho of -0.37. Thresholds for MPI and US$2.00 dailyincome per adult equivalent (adjusted to the World Bank’s private individual purchasing power parityestimate) are shown for reference. 56% of study households fall below the wealth threshold of US$2.00/adult/day, and 51% fall below the MPI threshold for classification as multidimensionally poor.




whereas participants from non-poor households were more likely to consult a qualified privatedoctor. Among children, patterns of utilization were similar for poor and non-poor partici-pants. Children from poor and non-poor groups were more likely to seek help from shops/pharmacies than adults.

The number of providers consulted was consistent between poor and non-poor participants(median of 2 providers before CMCH for both groups). 57% of poor and 63% of non-poor par-ticipants had sought help from two or more providers before attending CMCH (P = 0.254).19% of poor and 16% of non-poor participants sought help from three or more sources(P = 0.156), and 3% of poor and non-poor had sought help from four or more providers beforehospitalization at CMCH.

Timespan of pre-hospital care-seekingFig 4 Panels A through F illustrate the time taken to each of the three stages from onset ofsymptoms to arrival at CMCH for adults and children, and panels G and H show cumulativetimecourse from the complete duration of illness from onset of symptoms to arrival at CMCH.Wide variation was observed in the total time from onset of symptoms to arrival at the referralhospital.

The majority of participants sought help outside of the home soon after the onset of symp-toms. Among adults, MPI poor participants took a median of 10 hours longer than non-poorto decide to seek help outside the home (31 hours from onset for poor vs. 21 hours for non-poor; P = 0.049). The difference between poor and non-poor adults became more pronouncedfor the decision to escalate care to the referral hospital, with a difference in median time of 22hours (118 hours for poor vs. 96 hours for non-poor from onset to decision to escalate;P = 0.032). Once the decision to attend CMCH was made, few participants experienced majordelays in transport, and these delays affected poor and non-poor participants similarly, suchthat median total times from onset of symptoms to arrival at CMCH of 123 hours for pooradults vs. 101 hours for non-poor (P = 0.021).

Table 3. Summary of clinical diagnoses and deaths, disaggregated by age and MPI groups.


n = 527 (18 deaths) n = 269 (16 deaths) n = 258 (2 deaths)

Diagnostic Category n (%) Died n (%) Died n (%) Died P-valuea

Respiratory Tract Infection 110 (21%) . 56 (21%) . 54 (21%) . 0.351

Central Nervous System Infection 93 (18%) 11 61 (23%) 9 32 (12%) 2 0.002

Enteric Feverb 78 (15%) . 31 (12%) . 47 (18%) . 0.037

Urinary Tract Infection 55 (10%) 2 24 (9%) 2 31 (12%) . 0.258

Malaria 38 (7%) 3 28 (10%) 3 10 (4%) . 0.004

Dengue Feverb 34 (6%) . 10 (4%) . 24 (9%) . 0.012

Febrile Convulsion 23 (4%) . 12 (4%) . 11 (4%) . 1.000

Hepatobiliary Infection 23 (4%) . 12 (4%) . 11 (4%) . 1.000

Gastrointestinal Infection 10 (2%) 1 7 (3%) 1 3 (1%) . 0.340

Sepsis 9 (2%) 1 5 (2%) 1 4 (2%) . 1.000

Soft Tissue Infection 8 (2%) . 6 (2%) . 2 (1%) . 0.286

Undifferentiated Febrile Illness 46 (9%) . 17 (6%) . 29 (11%) . 0.063

aComparison of diagnostic category incidence for MPI poor vs. MPI non-poor participants, by Fisher’s exact test.bEnteric Fever and Dengue Fever were common clinical diagnoses, but could rarely be confirmed by microbiological/virological investigations, due to a

lack of laboratory resources.




Comparing children from poor and non-poor households, we found no evidence of differ-ence in care-seeking timecourses. Median times to first seeking help outside the home were 14and 16 hours for children from poor and non-poor households respectively, with no evidenceof difference (P = 0.749). Median times until escalation to the referral hospital were 93 hoursfrom onset for poor and 111 hours for non-poor children (P = 0.267), and median times toarrival were 97 and 119 hours for poor and non-poor children respectively (P = 0.395).

Using a multiple regression model, we analyzed demographic or geographical differencesbetween the MPI poor and non-poor groups to explain observed differences in referral time. Inaddition to MPI status the following variables were included in the model: sex, age, and journeyduration to hospital. Urban or rural classification was considered for the model, but droppedbecause of its strong correlation with distance to hospital. Pairwise correlation between theremaining variables confirmed no multicollinearity. In this model, there was no significantinteraction of MPI status and distance to hospital.

Fig 3. Patterns of care-seeking among participants prior to referral hospital admission. Patterns among MPI poor adults (A.), MPI non-poor adults (B.),MPI poor children (C.) and MPI non-poor children (D.). Arrows represent transitions between providers. Arrow sizes are proportional to the number ofparticipants making a transition between each pair of care providers, and ordered by frequency from top to bottom for the transitions between home and initialsource of help, and by frequency from left to right for transitions between sources of help consulted prior to the referral hospital. Transitions undertaken byfewer than 2% of the group are not illustrated.




Table 5 summarizes the multiple linear regression models for adults, children, and the com-bined population. For adults, MPI poor was associated with greater pre-hospital delays, inde-pendent of the other variables included in the model. The model indicates that, across thepopulation of adults with AFI, MPI poor patients face an additional 26 hours in the timespanfrom symptom onset to hospital arrival (95% CI: 7 to 46 hours) compared to non-poorpatients.

For adults, each hour of estimated travel time was associated with an additional 8.4 hours(95%CI: 1.5 to 15.4 hours) added to the total duration of illness before hospitalization. Theequivalent model for children yielded age as the only significant predictor of time from onsetto arrival, contributing 3.9 hours per year of age (95%CI: 1.9 to 6.0 hours). In the model foradults and children combined, distance from hospital was the only significant predictor of timefrom onset to arrival, with each hour of travel time associated with 6.3 hours added to the pre-hospital journey (95%CI: 1.4 to 11.1 hours).

Perceived delays in pre-hospital carePerceived contributors to delays in reaching hospital are summarized in Table 6. Thirteen per-cent of participants concluded that they had faced no delays in decision-making. Across bothMPI groups, the most prevalent perceived delay in decision-making was medical treatmentelsewhere, followed by uncertainty as to whether the patient was unwell enough to require hos-pitalization. These two responses were the only answers to significantly correlate with an

Table 4. Patterns of care-seeking among poor and non-poor adults and children.

Adult care-seeking pattern All Adults MPI Poor MPI Non-Poor Poor vs. Non-Poorn = 242 n = 98 n = 144

Source n (%) n (%) n (%) P-valuea

Shop/Pharmacy 79 (33%) 32 (33%) 47 (33%) 1.000

Private Doctor 97 (40%) 29 (30%) 68 (47%) 0.007

Private Allopath 60 (25%) 31 (32%) 29 (20%) 0.049

Government Health Complex/Clinic 45 (19%) 22 (22%) 23 (16%) 0.240

Government Hospital 41 (17%) 17 (17%) 24 (17%) 1.000

Traditional Healer 8 (3%) 4 (4%) 4 (3%) 0.718

Private Hospital 18 (7%) 9 (9%) 9 (6%) 0.458

Friends/Relatives 4 (2%) . . 4 (3%) 0.149

Other Source 2 (1%) 2 (2%) . . .

Child Care-seeking patterns All Children MPI Poor MPI Non-Poor Poor vs. Non-Poorn = 285 n = 171 n = 114

Source n (%) n (%) n (%) P-valuea

Shop/Pharmacy 187 (66%) 113 (66%) 74 (65%) 0.899

Private Doctor 129 (45%) 70 (41%) 59 (52%) 0.089

Private Allopath 85 (30%) 55 (32%) 30 (26%) 0.355

Government Health Complex/Clinic 81 (28%) 48 (28%) 33 (29%) 0.894

Government Hospital 35 (12%) 24 (12%) 11 (10%) 0.357

Traditional Healer 20 (7%) 12 (7%) 8 (7%) 1.000

Private Hospital 27 (9%) 15 (9%) 12 (11%) 0.692

Friends/Relatives 6 (2%) 5 (3%) 1 (1%) 0.407

Other Source 3 (1%) 2 (1%) 1 (1%) .

a Comparison of source among poor and non-poor participants by Fisher’s exact test.




Fig 4. Cumulative histograms of time to key stages in care-seeking. The total timespan from onset of symptoms to arrival at the referral hospital issubdivided into three stages: (A, B) time until the first decision to seek help outside the home; (C, D) time from this decision until escalation to a referralhospital; (E, F) time from the decision to escalate until arrival. Panels G and H represent the cumulative timecourse of all three stages for adults and childrenrespectively.


Table 5. Multiple linear regression analyses of pre-hospital illness timespan (in hours) for adults, children, and all participants.

Crude Coefficient (β)a 95% CI P-value Adjusted Coefficient (β)a 95% CI P-value

Adults:

MPI Poor 25.5 5.7 to 45.2 0.012 26.5 6.6 to 46.3 0.009

Distance to Hospital (hr) 8.9 1.9 to 15.9 0.013 8.4 1.5 to 15.4 0.017

Male 14.6 -5.4 to 34.5 0.152 17.8 -1.9 to 37.5 0.077

Age (yr) -0.5 -1.1 to 0.1 0.128 -0.62 -1.3 to 0.0 0.054

Children:MPI Poor -9.1 -28.5 to 10.2 0.354 -9.6 -28.9 to 9.7 0.327

Distance to Hospital (hr) 4.7 -2.0 to 11.4 0.170 5.9 -0.8 to 12.6 0.085

Male -3.2 -23.1 to 16.7 0.751 -3.3 -22.8 to 16.2 0.741

Age (yr) 3.9 1.9 to 5.9 <0.001 3.9 1.9 to 6.0 <0.001

Adults and Children:MPI Poor 6.2 -7.4 to 19.8 0.368 3.5 -10.4 to 17.4 0.625

Distance to Hospital (hr) 6.2 1.5 to 10.9 0.010 6.3 1.4 to 11.1 0.012

Male 5.1 -8.9 to 19.1 0.478 6.0 -8.1 to 20.1 0.402

Age (yr) 0.0 -0.4 to 0.4 0.977 0.1 -0.3 to 0.5 0.612

a The coefficient (β) reflects the magnitude (in hours) of the effect on the pre-hospital timespan associated with the variable’s presence (in the case of the

dichotomous variables MPI poor and male sex) or of each unit of the continuous variables, distance to hospital (in hours) or age (in years). Crude

(univariate) and adjusted (multivariate) coefficients are shown for each model. Variables associated with a statistically significant increase in pre-hospital

timespan in the multiple linear regression model are set in bold.




increase in total time from onset of symptoms to admission. MPI poor participants were morelikely to report delaying the decision to attend the referral hospital because of a lack of money.

The most commonly cited cause of delay in transport to hospital was the need to gatherfunds, reported by 42.3% of participants. MPI poor participants encountered this delay moreoften. MPI poor participants were also more likely to report that poor or busy roads had pro-longed their journeys. Means of transport are summarized in S3 Table.

Costs of pre-hospital treatment, and impoverishmentParticipants’ estimates of pre-hospital costs are shown in Table 7. In many cases, the memberof the household with greatest control over spending was unavailable for interview; onlyrespondents who were able to estimate costs are included in this analysis. Equivalent propor-tions of the poor and non-poor groups were able to estimate expenses (P = 0.164). In both poorand non-poor populations, expenditure was highest for investigations, medication, and medi-cal consumables. Poor participants spent significantly more than non-poor participants onaccommodation and transport. Total pre-hospital expenditure consumed a greater proportionof household income for poor participants.

Prior to hospital admission, a median of 7 days’ work was lost by poor households com-pared to 6 days’ work lost by non-poor households, (P = 0.012), with a strong skew towards asubset of patients in both groups requiring extensive care from multiple members of the

Table 6. Summary of perceived delays affecting the decision to attend referral hospital, and delays to transport after this decision.

MPI Poor MPI Non-Poor All Proportion test

n = 269 n = 258 n = 527

Perceived delay to decision to come to the hospital n % Of group n % of group n % of group P-value

Undergoing medical treatment elsewhere 170 (63%) 155 (60%) 325 (62%) 0.462

Unsure if unwell enough 108 (40%) 107 (41%) 215 (41%) 0.757

Not enough money 136 (51%) 60 (23%) 196 (37%) <0.001

Discussing decision within the family 72 (27%) 61 (24%) 133 (25%) 0.409

Undergoing traditional treatment or home remedies 20 (7%) 15 (6%) 35 (7%) 0.455

Concerned about time away from work/home 20 (7%) 10 (4%) 30 (6%) 0.078

Other delay (volunteered)a 8 (3%) 11 (4%) 19 (4%) –

No delay 37 (14%) 32 (12%) 69 (13%) 0.646

Perceived delays in reaching hospital after decision

Gathering funds 152 (57%) 71 (28%) 223 (42%) <0.001

Busy roads 101 (38%) 70 (27%) 171 (32%) 0.011

Arranging an escort 75 (28%) 67 (26%) 142 (27%) 0.621

Arranging a vehicle 63 (23%) 48 (19%) 111 (21%) 0.175

Poor roads 62 (23%) 30 (12%) 92 (17%) <0.001

Distance to hospital 54 (20%) 36 (14%) 90 (17%) 0.062

Too unwell to travel 9 (3%) 12 (5%) 21 (4%) 0.444

Frequent stops 9 (3%) 6 (2%) 15 (3%) 0.481

Slow/no vehicle 2 (1%) 1 (0%) 3 (1%) 0.587

Other delay (volunteered)b 7 (3%) 6 (2%) 13 (2%) –

No delay 62 (23%) 97 (36%) 159 (30%) <0.001

a Other sources of delays to the decision volunteered (number volunteering): negative impression of CMCH (6); positive impression of another source (6);

unfamiliar with choices for escalation (3); decision-maker unavailable (3); Ramadan (1); another household member unwell at home (1).b Other sources of delays to transport volunteered: night (8); unsure how to reach hospital (2); arranging leave from employer (1); no one available to care

for children (1).




household. This figure combined direct loss of the patient’s days of work and indirect loss bythose caring for him or her. The number of participating households that had already faced cat-astrophic costs by the time of hospital admission was estimated (Table 7). We found a highproportion of both poor and non-poor households had already experienced catastophic levelsexpenditure, with strong evidence of a greater effect on poorer households.

As a further assessment of economic consequences of pre-hospital management of AFI, par-ticipants were asked about their households’means of paying for the costs incurred. Thesources of payment are summarized in Table 8. The majority of participants in both MPI cate-gories were unable to pay from their household savings alone: overall, 78% had to seek fundsfrom outside of the household, and 63% undertook debts that would require repayment. MPIpoor participants were less likely to be able to pay from their own savings (P<0.001), and morelikely to take on debts to friends and family, banks, and other moneylenders (P<0.001).

DiscussionOur survey of patients with acute febrile illnesses identified important delays, obstacles, andcosts associated with multidimensional poverty, arising between onset of symptoms and arrivalat hospital. We found that for adults from multidimensionally poor households, median timeto arrival at the referral hospital was approximately one day longer than for those from non-poor households. This effect was not observed for children, among whom pre-hospital delays

Table 7. Estimated pre-hospital expenditure onmedical care, days lost from work by household members, and proportion of MPI groupsexperiencing catastrophic expenditure.

All MPI Poor MPI Non-Poor Poor vs.Non-Poor

Pre-hospital cost by type of expense(Tk)*

n Med [IQR, Range] n Med [IQR, Range] n Med [IQR, Range] P-valuea

Investigations 428 500 [0–1500,0–15000]

204 500 [0–1500,0–6000]

224 1000 [0–1500,0–15000]

0.610

Medication and medical consumables 353 900 [350–2000,0–25000]

175 900 [400–2000,0–10000]

178 900 [275–2000,10–25000]

0.441

Accommodation (patient and attendants) 421 0 [0–500,0–6730]

205 0 [0–606,0–6730]

216 0 [0–325,0–6500]

0.005

Transport 481 450 [200–850,0–8000]

246 500 [260–1000,0–8000]

235 350 [140–700,0–6200]

<0.001

Other known expenses 27 200 [200–500, 50–7000]

18 250 [100–500, 50–7000]

9 200 [200–400, 80–600]

0.979

Total pre-hospital costs

All expenses (absolute, Tk) 310 2575 [1050–5500,20–23500]

152 2750 [1280–5875,60–2150]

158 2500 [800–5180,20–23500]

0.097

All expenses (% of monthly income) 305 33% [12%-59%,0.1%-605%]

149 38% [22%-71%,1%-337%]

156 22% [7%-54%,0.1%-605%]

<0.001

Days of work lost by household 519 6 [4–12, 0–26] 266 7 [4–12, 0–26] 253 6 [3–10, 0–26] 0.012

Pre-hospital costs exceeding thresholdsfor catastrophic expenditure (n = 305)

n % n % n %

� 25% monthly household income 176 58% 101 68% 75 48% <0.001

� 40% monthly household income 131 43% 74 50% 57 37% 0.021

� 100% monthly household income 38 13% 24 16% 14 9% 0.059

*Includes estimates of 0Tk, but excludes those unable to offer an estimate for this category. Therefore, n varies between expense categories.a P-values obtained from Mann-Whitney U test for costs and proportion test for expenditure thresholds.




were of similar magnitude for poor and non-poor households. Given that timely initiation ofproperly directed antimicrobial therapy is a well-established determinant of survival in manysevere infectious diseases, this delay poses a substantial clinical risk [13–16]. We observed ahigher case fatality among patients from poorer households, and while a multitude of factorsundoubtedly contributed to this excess mortality, it underlines the urgency of reducing delayswherever possible.

Participants from all backgrounds recognized the need for help from outside the householdearly in the course of the AFI. However, even in this initial step, MPI poor adults took longerthan non-poor adults to seek help (with equivalent delays among poor and non-poor children).This implies that participants recognized the potential seriousness of AFI and its responsive-ness to treatment, but that barriers to care associated with poverty arise from an early stage.The interval from first seeking help to the household’s decision to escalate care to the referralhospital was longer for adults from poor households compared to non-poor. This may reflecthigher utilization of qualified private doctors by non-poor adults, compared to poor adults andchildren, who were more likely to consult unqualified practitioners. It is possible that qualifieddoctors were more likely to recognize the potential seriousness of AFI, and warn patients of thepossible need for escalation.

The differential effect of poverty on different age strata has more than one plausible explana-tion. Households may perceive childhood fever as more dangerous and urgent—and so prioritizethe timely care of children regardless of cost—reducing the impact of poverty on delays for chil-dren of both poor and non-poor households. Alternatively, non-poor adults may be prioritized,and encounter the fewest delays because their households are the most willing and able to pay.Both patterns of prioritization have been reported in care-seeking for acute febrile illnesses in dif-ferent contexts [37, 38]. Given the complexity of care-seeking behaviour, we hope to gain aclearer understanding of the decision-making process, and the differential effects of poverty atdifferent ages, through an ongoing qualitative study of this population. Reducing barriers to carestands to benefit all age strata, and may help to prevent the catastrophic levels of expenditureseen among febrile adults and children. Therefore, we would recommend addressing barriers tocare for all ages, though the greatest gains of equity may be seen among adults.

Table 8. Sources of payment for expenses arising from illness before arrival at the referral hospital.


n = 527 n = 269 n = 258

Means of payment for pre-hospital care n (%) n (%) n (%) P-valuea

Own Savings 376 (71%) 171 (64%) 205 (79%) <0.001

Loan from Relatives/Friends 316 (60%) 200 (74%) 116 (45%) <0.001

Gift from Relatives/Friends 124 (24%) 70 (26%) 54 (21%) 0.168

Other Moneylender 92 (17%) 68 (25%) 24 (9%) <0.001

Bank Loans 13 (2%) 13 (5%) – <0.001

Sale of Propertyb 9 (2%) 6 (2%) 3 (1%) 0.344

Other Sourcec 4 (1%) – 4 (2%) –

Liquidity of payment

Unable to pay from own savings alone 410 (78%) 246 (91%) 164 (64%) <0.001

Unable to pay without incurring debts 334 (63%) 213 (79%) 121 (47%) <0.001

aP-values from proportion tests, Poor vs. Non-Poor.b Property sold: land (2); livestock (2); tea shop (1); rickshaw (1); tree (1); gold ornaments (1); television (1).c Other sources of payment volunteered: payment by employer (4).




Our findings suggest that the best opportunity to reduce pre-hospital delays is during theperiod in which patients are seeking help outside the home, before the decision to escalate tothe referral hospital is made. It is important, therefore, to analyze where healthcare is sought,and what routes patients follow when an illness progresses. In keeping with international stud-ies, most patients followed a pattern of escalation from sources nearest the home, most accessi-ble, and least expensive, toward greater professional expertise and more intensive management[33]. We observed few consultations with non-allopathic traditional healers, either by MPIpoor or non-poor households. This corroborates previous observations that the population rec-ognizes AFIs as amenable to allopathic treatment [18, 28, 39].

In most cases, a private sector hierarchy—from shop to private allopath/doctor to hospital—supplanted the public sector hierarchy from clinic to health complex to hospital. Despiteincreased government investment in public-sector healthcare, serial community surveys haveshown falling perception of quality and satisfaction, with highest dissatisfaction among the poor-est households [28]. Patients have identified under-staffing, poor physical facilities, behavior ofservice providers, and inability to provide essential medicines directly as causes of this low opin-ion. Healthcare providers have noted consistent problems with drug procurement and supply.Absenteeism is common—reported at 40% for doctors at health complexes nationwide, risinghigher in poorer areas [40, 41]. The majority of patients attending public sector services arerequired to pay for medicines from private vendors due to depletion of essential drug stocks, andmany experience pressure to pay unofficial fees to members of staff [28, 42–44].

Participants perceived financial constraints as a major cause of delays in seeking healthcare.These constraints disproportionately affected poorer households, but were a problem for lesspoor households as well. AFI can be financially disastrous to households already afflicted bypoverty, and that unmanageable costs begin to accrue early in the patient journey. At the pointof arrival at the referral hospital, a majority of households had already extended beyond theirsavings to pay for medical care, incurred debts that would require repayment, and reached thepoint of catastrophic expenditure. All of these consequences were more common for poorerhouseholds. Expenses usually rise steeply after admission, and families face the added burdensof time away from work and accommodation far from home for those who accompany thepatient [32]. The decision to seek medical care poses a considerable risk of financial ruin, par-ticularly for those already afflicted by multidimensional poverty [32, 45]. Addressing this riskmay reduce delays due to reluctance to take on this burden, and those due to the need to gatherfunds.

Several constructive recommendations for the private and public sectors can be drawn fromthis investigation. Our findings imply that there is scope to improve the accessibility and qualityof the government sector, so that its well-structured but under-utilized hierarchy becomes adesirable option for patients with AFI. Addressing absenteeism, ensuring that patients are awareof their rights to free essential drugs and services, and extending government subsidies to the dis-pensation of more drugs and ancillary care will improve both the quality and the uptake of publicsector facilities closer to home, and reduce costs and delays for patients [28, 41].

Educating informal private sector providers in the immediate management of AFIs—partic-ularly in malaria-endemic parts of the Division—might improve such practices, and save lives[10, 22]. Engagement initiatives should offer education on the recognition of warning signs(such as reduced level of consciousness) that should prompt a provider to refer a patient to aqualified doctor or hospital. New tools such as RDTs for malaria and other infections have thepotential to give para-professional providers a limited but important role in diagnosis andtreatment algorithms. This could help to integrate public and private services, and streamlinewhat is currently a long and costly process of serial consultation. As the first responders tomany potentially fatal illnesses, providers in this sector have a critical opportunity to intervene.



Provision of insurance schemes to the poor should be promoted. Access to health insurancewill encourage early medical consultation with qualified practitioners—without the need togather funds and take on debts. Insurance also increases capacity to respond to high costs ofin-patient care when this is necessary [46, 47]. Demand-side financing has been introduced topromote maternal health in rural Bangladesh, by providing credit vouchers in advance of needfor health services. This appears to reduce cost and increase consultations with appropriatepractitioners [48]. Extending such programs may help to reduce household expenditure andpromote early consultation and escalation. Previous studies of insurance schemes in Bangla-desh have identified benefits in terms of access to basic and preventative healthcare, but foundthat insured households are not protected from the catastrophic expenses of severe illness,necessitating hospital admission, where the costs of interventions and supportive care are notcovered, and can escalate steeply [47]. Extending such programmes to cover high-cost carewould help to mitigate the financial ruin that can stem from acute illness.

In a setting where hospital admissions pose considerable financial strain, and where in-patient demand exceeds provision, recommendations that leads to too many hospital admis-sions could be as dangerous as those that lead to too few [32]. An optimal community servicewould prevent a majority of hospital admissions by providing diagnosis and treatment early,while fast-tracking those cases that require urgent escalation.

In the long term, social changes that address inequities in assets, living standards, education,and chronic health in Bangladesh will have positive consequences for the pre-hospital manage-ment of AFIs and other causes of preventable morbidity and mortality. There is evidence fromlongitudinal studies that integrated development interventions have already had a positiveimpact on health seeking as well as health outcomes elsewhere in Bangladesh [49, 50].

Our study has several important limitations. The potential for selection bias is acknowledged.The goal of daily recruitment of one patient from the adult medical wards and one from the paedi-atric medical wards was aimed at ensuring a systematic sample through the year, but more consid-eration could have been given to ensuring a random selection among those febrile patientsadmitted during a 24-hour period. Failure to randomize among each day’s admissions may haveled to under-sampling of participants who were admitted for a short period, with rapid dischargeor clinical deterioration and death. It is likely that such under-sampling would be non-differential,affecting poor and non-poor groups equally; this would bias the results toward the null hypothesis,and lead to a more conservative estimate of difference between poor and non-poor populations.

This study does not capture those febrile illnesses that did not lead to hospitalization—either because adequate care was obtained elsewhere, or because barriers to hospital care wereinsurmountable. It is likely that this study under-represents those residents of the ChittagongDivision who are worst afflicted by multidimensional poverty, as well as those who are geo-graphically most remote, since these groups might never reach the referral hospital.

The potential for misclassification bias is also acknowledged, with regard to ascertainmentof MPI status in the context of acute illness. While all other components of MPI can be ascer-tained based on household characteristics preceding the illness, nutrition status of the house-hold must be measured at the time of interview, and could potentially be affected by rapidweight loss as a consequence of AFI. This could have led to over-estimation of malnutritionand misclassification of some non-poor households as poor. There were 39 cases in which theparticipant’s low nutritional status could have affected MPI sufficiently to change classification.While it is unlikely that all patients experienced marked weight loss, in this scenario the maxi-mum rate of misclassification due to acute weight loss was 7%. Had these cases been re-classi-fied, the assessment of the primary outcome measures would have been unchanged.

Acknowledging these caveats, our study contributes to a more complete understanding ofcare-seeking behavior and the obstacles posed by poverty. This understanding is essential to



confronting the complex relationship between poverty and illness, and prioritizing interven-tions that address these two inseparable problems.

Supporting InformationS1 Table. Transitions between sources of healthcare undertaken by participants with AFI.(DOCX)

S2 Table. Routes to the referral hospital, undertaken by 527 patients with AFI, beginningwith the first source of help outside the home, and ranked for frequency of utilization.(DOCX)

S3 Table. Means of transport to the referral hospital.(DOCX)

AcknowledgmentsWe thank our Clinical Research Team, Dr Zabeen Chowdhury, Dr Aparup Kanit Das, DrKonika Dey, Dr Nafiz Iqbal, Dr Regina Islam, and Dr. Mizanur Rahman, for their exemplarywork. We thank Prof. Md Abul Faiz for his advice and insight into healthcare in the Division.

Author ContributionsConceived and designed the experiments: MTH RJM RS RK ADMAH. Performed the experi-ments: MTHMSC HWFK. Analyzed the data: MTH RJM AJ. Contributed reagents/materials/analysis tools: MTHMSC. Wrote the paper: MTH RJM HWFK RS RK ADMAH.

References1. Gwatkin DR, Rutstein S, Johnson K, Suliman E, Wagstaff A, Amouzou A. Socio-economic differences

in health, nutrition, and population within developing countries: an overview. Washington, DC: TheWorld Bank, 2007.

2. Oxford Poverty and Human Development Initiative. "Bangladesh Country Briefing" Available at: http://www.ophi.org.uk/multidimensional-poverty-index/mpi-country-briefings/bangladesh: OPHI, Universityof Oxford; 2013 [cited 2013 15 February].

3. Sen B. Health and poverty in the context of country development strategy: a case study on Bangladesh.Geneva: W.H.O.; 1997.

4. Ahmed SM, Tomson G, Petzold M, Kabir ZN. Socioeconomic status overrides age and gender in deter-mining health-seeking behaviour in rural Bangladesh. Bulletin of the World Health Organization. 2005;83(2):109–17. doi: /S0042-96862005000200011. PMID: 15744403; PubMed Central PMCID:PMC2623805.

5. Wagstaff A. Poverty and health sector inequalities. Bulletin of theWorld Health Organization. 2002; 80(2):97–105. PMID: 11953787; PubMed Central PMCID: PMC2567730.

6. Dhillon PK, Jeemon P, Arora NK, Mathur P, Maskey M, Sukirna RD, et al. Status of epidemiology in theWHO South-East Asia region: burden of disease, determinants of health and epidemiological research,workforce and training capacity. International journal of epidemiology. 2012; 41(3):847–60. doi: 10.1093/ije/dys046 PMID: 22617689; PubMed Central PMCID: PMC3396314.

7. Victora CG, Wagstaff A, Schellenberg JA, Gwatkin D, Claeson M, Habicht JP. Applying an equity lensto child health and mortality: more of the same is not enough. Lancet. 2003; 362(9379):233–41. doi: 10.1016/S0140-6736(03)13917-7 PMID: 12885488.

8. Health WCoSDo. Closing the Gap in a Generation Health Equity through Action on the Social Determi-nants of Health. Geneva: World Health Organization; 2008. Available from: http://www.gallaudet.eblib.com/patron/FullRecord.aspx?p=361775.

9. World Health Organization. WHO Bangladesh Country Cooperation Strategy 2008–2013. Geneva:WHO, 2007.



http://www.plosone.org/article/fetchSingleRepresentation.action?uri=info:doi/10.1371/journal.pone.0152965.s001



http://www.ophi.org.uk/multidimensional-poverty-index/mpi-country-briefings/bangladesh:

http://www.ophi.org.uk/multidimensional-poverty-index/mpi-country-briefings/bangladesh:

http://www.ncbi.nlm.nih.gov/pubmed/15744403


http://dx.doi.org/10.1093/ije/dys046

http://dx.doi.org/10.1093/ije/dys046


http://dx.doi.org/10.1016/S0140-6736(03)13917-7

http://dx.doi.org/10.1016/S0140-6736(03)13917-7


http://www.gallaudet.eblib.com/patron/FullRecord.aspx?p=361775

http://www.gallaudet.eblib.com/patron/FullRecord.aspx?p=361775

10. Ahmed SM. Differing health and health-seeking behaviour: ethnic minorities of the Chittagong HillTracts, Bangladesh. Asia-Pacific journal of public health / Asia-Pacific Academic Consortium for PublicHealth. 2001; 13(2):100–8. PMID: 12597507.

11. Najnin N, Bennett CM, Luby SP. Inequalities in care-seeking for febrile illness of under-five children inurban Dhaka, Bangladesh. Journal of health, population, and nutrition. 2011; 29(5):523–31. PMID:22106759; PubMed Central PMCID: PMC3225115.

12. Acestor N, Cooksey R, Newton PN, Menard D, Guerin PJ, Nakagawa J, et al. Mapping the aetiology ofnon-malarial febrile illness in Southeast Asia through a systematic review—terra incognita impairingtreatment policies. PloS one. 2012; 7(9):e44269. doi: 10.1371/journal.pone.0044269 PMID: 22970193;PubMed Central PMCID: PMC3435412.

13. Lin AL, Safdieh JE. The evaluation and management of bacterial meningitis: current practice andemerging developments. The neurologist. 2010; 16(3):143–51. doi: 10.1097/NRL.0b013e3181d14185PMID: 20445424.

14. Kumar A, Roberts D, Wood KE, Light B, Parrillo JE, Sharma S, et al. Duration of hypotension before ini-tiation of effective antimicrobial therapy is the critical determinant of survival in human septic shock.Critical care medicine. 2006; 34(6):1589–96. doi: 10.1097/01.CCM.0000217961.75225.E9 PMID:16625125.

15. Gomes MF, Faiz MA, Gyapong JO, WarsameM, Agbenyega T, Babiker A, et al. Pre-referral rectal arte-sunate to prevent death and disability in severe malaria: a placebo-controlled trial. Lancet. 2009; 373(9663):557–66. doi: 10.1016/S0140-6736(08)61734-1 PMID: 19059639; PubMed Central PMCID:PMC2646124.

16. Rivers E, Nguyen B, Havstad S, Ressler J, Muzzin A, Knoblich B, et al. Early goal-directed therapy inthe treatment of severe sepsis and septic shock. The New England journal of medicine. 2001; 345(19):1368–77. doi: 10.1056/NEJMoa010307 PMID: 11794169.

17. Tusting LS, Willey B, Lucas H, Thompson J, Kafy HT, Smith R, et al. Socioeconomic development asan intervention against malaria: a systematic review and meta-analysis. Lancet. 2013. doi: 10.1016/S0140-6736(13)60851-X PMID: 23790353.

18. El Arifeen S, Baqui AH, Victora CG, Black RE, Bryce J, Hoque DM, et al. Sex and socioeconomic differ-entials in child health in rural Bangladesh: findings from a baseline survey for evaluating IntegratedManagement of Childhood Illness. Journal of health, population, and nutrition. 2008; 26(1):22–35.PMID: 18637525; PubMed Central PMCID: PMC2740675.

19. Bhuiya A, Mahmood SS, Rana AK, Wahed T, Ahmed SM, Chowdhury AM. A multidimensionalapproach to measure poverty in rural Bangladesh. Journal of health, population, and nutrition. 2007; 25(2):134–45. PMID: 17985815; PubMed Central PMCID: PMC2753999.

20. Alkire S, Santos ME. OPHI Working Paper No. 38 Acute Multidimensional Poverty: A New Index forDeveloping Countries. Oxford, UK: Oxford Poverty & Human Development Initiative, 2010.

21. United Nations Development Programme. Human Development Report 2010 The Real Wealth ofNations: Pathways to Human Development. New York: United Nations Development Programme,2010.

22. Ashraf A, Chowdhury S, Streefland P. Health, disease and health-care in rural Bangladesh. Social sci-ence & medicine. 1982; 16(23):2041–54. PMID: 7157035.

23. Pitchforth E, van Teijlingen E, GrahamW, Fitzmaurice A. Development of a proxy wealth index forwomen utilizing emergency obstetric care in Bangladesh. Health policy and planning. 2007; 22(5):311–9. doi: 10.1093/heapol/czm022 PMID: 17698888.

24. Killewo J, Anwar I, Bashir I, Yunus M, Chakraborty J. Perceived delay in healthcare-seeking for epi-sodes of serious illness and its implications for safe motherhood interventions in rural Bangladesh.Journal of health, population, and nutrition. 2006; 24(4):403–12. PMID: 17591337; PubMed CentralPMCID: PMC3001144.

25. Bangladesh Bureau of Statistics. Population and Housing Census 2011: Socio-economic and Demo-graphic Report, National Series. 2012;4.

26. Sayem ASM. Best Practices & Alternative Integrated Community Based Model in Delivering PrimaryHealth Care Service for Chittagong Hill Tracts of Bangladesh. [MIH Dissertation]. In press 2011.

27. Ahmed SM, Evans TG, Standing H, Mahmud S. Harnessing pluralism for better health in Bangladesh.Lancet. 2013; 382(9906):1746–55. doi: 10.1016/S0140-6736(13)62147-9 PMID: 24268003.

28. Cockroft A, Milne D, Andersson N. Bangladesh Health and Population Sector Programme 1998–2003:The Third Service Delivery Survey Final Report. Dhaka: CIETcanada and Ministry of Health and Fam-ily Welfare, Government of the People's Republic of Bangladesh, 2004.





http://dx.doi.org/10.1371/journal.pone.0044269


http://dx.doi.org/10.1097/NRL.0b013e3181d14185


http://dx.doi.org/10.1097/01.CCM.0000217961.75225.E9


http://dx.doi.org/10.1016/S0140-6736(08)61734-1


http://dx.doi.org/10.1056/NEJMoa010307


http://dx.doi.org/10.1016/S0140-6736(13)60851-X

http://dx.doi.org/10.1016/S0140-6736(13)60851-X





http://dx.doi.org/10.1093/heapol/czm022



http://dx.doi.org/10.1016/S0140-6736(13)62147-9


29. National Institute of Population Research and Training (NIPORT), Mitra and Associates, Macro Interna-tional. Bangladesh Demographic and Health Survey 2007. Dhaka, Bangladesh: National Institute ofPopulation Research and Training, Mitra and Associates, and Macro International., 2009.

30. Organisation for Economic Co-operation and Development. The OECD list of social indicators. Paris:Organisation for Economic Co-operation and Development; 1982. 124 p. p.

31. World Bank Development Research Group. "Poverty headcount ratio at $1.25 a day (PPP) (% of popu-lation)" 2013 [cited 2013 26/7/2013]. Available from: http://data.worldbank.org/indicator/SI.POV.DDAY.

32. Rahman MM, Gilmour S, Saito E, Sultana P, Shibuya K. Health-related financial catastrophe, inequalityand chronic illness in Bangladesh. PloS one. 2013; 8(2):e56873. doi: 10.1371/journal.pone.0056873PMID: 23451102; PubMed Central PMCID: PMC3581555.

33. Ryan GW.What do sequential behavioral patterns suggest about the medical decision-making pro-cess?: modeling home case management of acute illnesses in a rural Cameroonian village. Social sci-ence & medicine. 1998; 46(2):209–25. PMID: 9447644.

34. Ferro-Luzzi A, JamesWP. Adult malnutrition: simple assessment techniques for use in emergencies.The British journal of nutrition. 1996; 75(1):3–10. PMID: 8785188.

35. Webb P, Bhatia R. A Manual: Measuring and Interpreting Malnutrition and Mortality. In: ProgrammeWF, editor. Rome2005.

36. Bangladesh Bureau of Statistics. Report on Sample Vital Registration System-2010. Dhaka, Bangladesh:Statistics Division, Ministry of Planning, Government of the People's Republic of Bangladesh, 2011.

37. Elsinga J, Lizarazo EF, Vincenti MF, Schmidt M, Velasco-Salas ZI, Arias L, et al. Health SeekingBehaviour and Treatment Intentions of Dengue and Fever: A Household Survey of Children and Adultsin Venezuela. PLoS Negl Trop Dis. 2015; 9(12):e0004237. doi: 10.1371/journal.pntd.0004237 PMID:26624283; PubMed Central PMCID: PMCPMC4666462.

38. Vialle-Valentin CE, LeCates RF, Zhang F, Ross-Degnan D. Treatment of Febrile illness with artemisinincombination therapy: prevalence and predictors in five African household surveys. J Pharm PolicyPract. 2015; 8(1):1. doi: 10.1186/s40545-014-0024-0 PMID: 25926989; PubMed Central PMCID:PMCPMC4403946.

39. Ahmed SM, Haque R, Haque U, Hossain A. Knowledge on the transmission, prevention and treatmentof malaria among two endemic populations of Bangladesh and their health-seeking behaviour. Malariajournal. 2009; 8:173. doi: 10.1186/1475-2875-8-173 PMID: 19640282; PubMed Central PMCID:PMC2729311.

40. Chaudhury N, Hammer J, Kremer M, Muralidharan K, Rogers FH. Missing in action: teacher and healthworker absence in developing countries. The journal of economic perspectives: a journal of the Ameri-can Economic Association. 2006; 20(1):91–116. PMID: 17162836.

41. Chaudhury N, Hammer J. Ghost Doctors: Absenteeism in Bangladeshi Health Facilities. PolicyResearchWorking Paper 3065. TheWorld Bank Development Research Group, 2003.

42. Afsana K. The tremendous cost of seeking hospital obstetric care in Bangladesh. Reproductive healthmatters. 2004; 12(24):171–80. PMID: 15626207.

43. Chowdhury Z. The politics of essential drugs: the makings of a successful health strategy: lessons fromBangladesh. London: Zed; 1995.

44. Killingsworth JR, Hossain N, Hedrick-Wong Y, Thomas SD, Rahman A, Begum T. Unofficial fees inBangladesh: price, equity and institutional issues. Health policy and planning. 1999; 14(2):152–63.PMID: 10538718.

45. Dror DM, van Putten-Rademaker O, Koren R. Cost of illness: evidence from a study in five resource-poor locations in India. The Indian journal of medical research. 2008; 127(4):347–61. PMID: 18577789.

46. Knaul FM, Arreola-Ornelas H, Mendez-Carniado O, Bryson-Cahn C, Barofsky J, Maguire R, et al. Evi-dence is good for your health system: policy reform to remedy catastrophic and impoverishing healthspending in Mexico. Lancet. 2006; 368(9549):1828–41. doi: 10.1016/S0140-6736(06)69565-2 PMID:17113432.

47. Werner WJ. Micro-insurance in Bangladesh: risk protection for the poor? Journal of health, population,and nutrition. 2009; 27(4):563–73. PMID: 19761089; PubMed Central PMCID: PMC2928102.

48. Ahmed S, Khan MM. A maternal health voucher scheme: what have we learned from the demand-sidefinancing scheme in Bangladesh? Health policy and planning. 2011; 26(1):25–32. doi: 10.1093/heapol/czq015 PMID: 20375105.

49. Ahmed SM, Adams AM, Chowdhury M, Bhuiya A. Changing health-seeking behaviour in Matlab, Ban-gladesh: do development interventions matter? Health policy and planning. 2003; 18(3):306–15. PMID:12917272.

50. Ahmed SM, Adams AM, Chowdhury M, Bhuiya A. Gender, socioeconomic development and health-seeking behaviour in Bangladesh. Social science & medicine. 2000; 51(3):361–71. PMID: 10855923.



http://data.worldbank.org/indicator/SI.POV.DDAY

http://dx.doi.org/10.1371/journal.pone.0056873




http://dx.doi.org/10.1371/journal.pntd.0004237


http://dx.doi.org/10.1186/s40545-014-0024-0


http://dx.doi.org/10.1186/1475-2875-8-173






http://dx.doi.org/10.1016/S0140-6736(06)69565-2



http://dx.doi.org/10.1093/heapol/czq015

http://dx.doi.org/10.1093/heapol/czq015




Documents

The Relationship between Poverty and Healthcare Seeking ... · Charles Darwin University The Relationship between Poverty and Healthcare Seeking among Patients Hospitalized with Acute